Databricks DA Assoc · Domain 2 of 9

Databricks DA Assoc Domain 2: Managing Data

Section 2 is about the day-to-day handling of governed data: finding and using certified datasets in Unity Catalog, tagging assets with governed tags, following lineage, and cleaning tables with SQL.

Exam weight
8%
Official objectives
3
Estimated questions estimate
~4
Suggested study time estimate
4–6 hrs

About 11 min read Reviewed October 2026

Overview

Managing Data is 8% of the exam and has three objectives. The first is discovery: using Unity Catalog search, Catalog Explorer and certification signals to locate the right dataset, query it, and manage it as an owner. The second covers metadata. You apply governed tags to a table or column and read lineage, now including external lineage that connects Unity Catalog assets to systems outside Databricks such as BI tools or source applications. The third objective is hands-on SQL cleaning of Unity Catalog tables: filtering out invalid rows, deduplicating, standardising values and handling NULLs with functions like COALESCE, NULLIF and TRY_CAST.

Governed tags and external lineage are new wording in the October 2026 guide. Governed tags are defined at account level with policies on allowed values and on who may assign them, which is what separates them from free-form tags. The cleaning objective is practical; expect to read SQL and predict its effect.

How it's tested

Questions mix short UI scenarios with SQL snippets. You may be asked where to apply a tag, what lineage shows, or what a cleaning statement returns. Read SQL carefully for NULL handling and for whether the statement changes the table or only the result set.

  • Find the trusted source

    Several similar tables exist and you choose how to identify the certified, owner-endorsed dataset through Unity Catalog search and Catalog Explorer.

  • Tag and trace

    A governance request asks you to label sensitive columns with a controlled value or to see which dashboards depend on a table.

  • Predict the cleaned output

    A SQL statement using COALESCE, NULLIF, TRY_CAST, DISTINCT or a WHERE filter is shown, and you predict the rows or values that remain.

Reasoning it through

A query must replace missing discount values with zero but leave real zeros and negative adjustments untouched. Filtering WHERE discount IS NOT NULL drops rows, which loses orders. CASE WHEN discount = 0 handles the wrong values. COALESCE(discount, 0) returns the first non-NULL argument, so only missing values become zero while every existing value is preserved, which is what the requirement asks for.

Objectives 3

Official objectives from the Databricks outline, in outline order.

2.1 Use Unity Catalog to discover, query, and manage certified datasets.Locating and using trusted, governed data.
What it means
Unity Catalog search and Catalog Explorer let you find tables by name, comment, tag or column. Certified datasets carry an endorsement that marks them as the approved source. Once found, you query them by their three-level name and, as an owner, maintain comments, ownership and permissions.
What to understand
Discovery depends on good metadata: table and column comments, tags and certification status make search results useful, and AI-generated comments can fill gaps. You only see objects you have some privilege on, and querying needs USE CATALOG, USE SCHEMA and SELECT. Managing a dataset includes editing its description, setting the owner, and deprecating outdated copies so users settle on the certified version.
How it shows up
Expect a scenario with duplicate-looking tables where the right action is to use the certified one or to certify it.
Common confusion
Believing certification copies the table or changes its permissions.
2.2 Use the Catalog Explorer to tag a data asset with governed tags and view its lineage, including external lineage.Governed tags, lineage graphs and external lineage.
What it means
Tags are key-value labels on catalogs, schemas, tables, columns and other objects. Governed tags are defined centrally with a policy that limits allowed values and who can apply them. The Lineage tab shows how data flows between tables, columns, notebooks, jobs, pipelines and dashboards, and external lineage adds systems outside Databricks.
What to understand
Governed tags keep classification consistent, for example pii with values such as email or phone, and they can be referenced by attribute-based access control policies. Free-form tags have no such guardrails. Unity Catalog records lineage at table and column level from queries it executes, without manual work. External lineage lets you register upstream sources or downstream consumers, such as a Salesforce source or a Tableau workbook, so impact analysis reaches beyond the workspace.
How it shows up
A request to label sensitive columns consistently, or to find every downstream consumer before changing a table, is the typical stem.
Common confusion
Thinking lineage covers external tools automatically; those links must be registered as external metadata.
2.3 Perform data cleaning on Unity Catalog Tables in SQL, including removing invalid data or handling missing values.SQL techniques for removing bad rows and filling gaps.
What it means
Cleaning means making a table fit for analysis: removing duplicates and invalid records, standardising formats, converting types safely, and dealing with missing values either by filtering them, replacing them with defaults or flagging them.
What to understand
Key tools are WHERE filters, DISTINCT or ROW_NUMBER for deduplication, TRIM, UPPER and LOWER for standardising text, TRY_CAST to turn bad strings into NULL instead of failing, COALESCE and IFNULL to supply defaults, NULLIF to turn sentinel values like empty strings into NULL, and DELETE or UPDATE to change a Delta table in place. Remember NULL comparisons return NULL, so use IS NULL rather than = NULL, and COUNT(column) skips NULLs while COUNT(*) does not.
How it shows up
Questions show a cleaning statement and ask for its result, or ask which function meets a stated rule for missing or invalid values.
Common confusion
Writing WHERE col = NULL, which matches nothing, instead of WHERE col IS NULL.

High-yield concepts

The terms, rules and formulas this domain turns into questions most often.

  • Certified dataset

    Unity Catalog asset endorsed as the trusted version.

    Remember
    Prefer it over look-alike copies.
    Tested as
    Discovery scenarios.
    Trap
    Assuming it grants SELECT.

  • Governed tag

    Tag whose key, allowed values and assigners are controlled by policy.

    Remember
    Consistent classification across assets.
    Tested as
    New wording in the 2026 guide.
    Trap
    Treating it like a free-form tag.

  • Attribute-based access control

    Policies that act on tagged objects.

    Remember
    Governed tags can drive masking and filtering.
    Tested as
    Links tagging to security.
    Trap
    Thinking tags alone restrict access.

  • Lineage tab

    Graph of upstream and downstream assets.

    Remember
    Captured automatically, down to columns.
    Tested as
    Impact analysis.
    Trap
    Believing it needs manual entry.

  • External lineage

    Lineage links to systems outside Databricks.

    Remember
    Registered as external metadata.
    Tested as
    Impact beyond the workspace.
    Trap
    Expecting automatic capture from BI tools.

  • COALESCE

    Returns the first non-NULL argument.

    Remember
    Fills missing values without dropping rows.
    Tested as
    Missing value handling.
    Trap
    Using it to fix invalid non-NULL values.

  • NULLIF

    Returns NULL when two values are equal.

    Remember
    Turns sentinels like '' into NULL.
    Tested as
    Standardising missing markers.
    Trap
    Reversing it with COALESCE.

  • TRY_CAST

    Converts types, returning NULL on failure.

    Remember
    Avoids query errors on bad strings.
    Tested as
    Invalid data handling.
    Trap
    Expecting it to raise an error.

  • ROW_NUMBER deduplication

    Window function to keep one row per key.

    Remember
    Filter where row number equals 1.
    Tested as
    Removing duplicates by key.
    Trap
    Using DISTINCT when rows differ in some columns.

  • IS NULL

    Correct test for missing values.

    Remember
    = NULL never evaluates true.
    Tested as
    Filter logic.
    Trap
    Writing = NULL.

Side by side

Concepts candidates mix up, compared on the point that decides the answer.

Tags and lineage

ConceptMeaningKey differenceCommon trap
Governed tag Centrally defined key with allowed values. Enforced vocabulary and assignment rights. Assuming any user can create values.
Free-form tag Ad hoc key-value label. No policy or value control. Relying on it for compliance.
Lineage Automatically captured data flow inside Unity Catalog. Covers tables, columns, notebooks, jobs and dashboards. Thinking it is manual.
External lineage Registered links to outside systems. Must be added through external metadata. Expecting it to appear by default.

Missing and invalid value functions

ConceptMeaningKey differenceCommon trap
COALESCE(a, b) First non-NULL value. Fills gaps. Expecting it to change empty strings.
NULLIF(a, b) NULL if a equals b. Creates NULLs from sentinels. Confusing it with IFNULL.
TRY_CAST(x AS type) Safe conversion. Bad input becomes NULL. Using CAST and getting errors.
DELETE FROM ... WHERE Removes rows from a Delta table. Changes the stored table. Thinking a SELECT filter alters the table.

Exam traps

Each trap is an answer that looks right. Learn the clue that should change your mind.

  1. Filtering instead of filling

    What candidates pick
    WHERE value IS NOT NULL to handle missing values.
    Why it looks right
    It removes the problem quickly.
    Why it's wrong
    Filtering throws away whole records; when the rule says keep rows and supply a default, COALESCE is required.

    Clue to watch for Keep all rows, replace missing values.

  2. NULL equality

    What candidates pick
    WHERE region = NULL.
    Why it looks right
    It reads naturally.
    Why it's wrong
    Any comparison with NULL yields NULL, never true, so no rows match; IS NULL is the correct predicate.

    Clue to watch for Rows with missing values are not returned.

  3. CAST on dirty data

    What candidates pick
    CAST(amount AS DECIMAL).
    Why it looks right
    It is the standard conversion.
    Why it's wrong
    With ANSI mode on, CAST fails the whole query on a bad string, whereas TRY_CAST returns NULL for that value.

    Clue to watch for Some values contain text like 'N/A'.

  4. Free-form tag for compliance

    What candidates pick
    Add a custom tag named sensitive.
    Why it looks right
    Tags are quick to add.
    Why it's wrong
    Without a governed tag policy, values and assigners are uncontrolled, so classification drifts and policies cannot rely on it.

    Clue to watch for Consistent, controlled classification.

  5. Lineage for Tableau

    What candidates pick
    Lineage is automatic for every consumer.
    Why it looks right
    Unity Catalog lineage is automatic internally.
    Why it's wrong
    Tools outside Databricks are only shown when registered as external lineage metadata.

    Clue to watch for Downstream reports built outside the workspace.

  6. Certification as access

    What candidates pick
    Certify the table so analysts can read it.
    Why it looks right
    Certification sounds like approval.
    Why it's wrong
    Certification signals trust; analysts still need USE CATALOG, USE SCHEMA and SELECT privileges.

    Clue to watch for Users can find the table but cannot query it.

Answer framework

Apply this sequence to discovery, tagging, lineage and cleaning questions in this section.

  1. Decide whether the question is about finding data, labelling it, tracing it or fixing its values.
  2. For discovery, prefer certified assets found through Unity Catalog search, and remember access still requires grants. Certification signals trust; it does not grant SELECT.
  3. For labelling, choose governed tags when consistency or policies matter, and free-form tags only for informal notes.
  4. For tracing, use the Lineage tab inside Databricks and external lineage for systems outside it. Lineage is captured automatically for queries that run against Unity Catalog tables.
  5. For cleaning, decide whether to drop, fill, convert or deduplicate, then pick WHERE, COALESCE, TRY_CAST or ROW_NUMBER. TRY_CAST returns NULL instead of failing on bad values, which keeps a load running.

Practice question

An analyst must convert a string column order_total to a decimal. Some rows contain values like 'unknown'. The query must not fail, and invalid values should later be replaced with 0. Which expression meets both requirements?

Want more like this? The Databricks DA Assoc practice test has questions for every domain.

Checklist

0 of 10

Tick each one you can do without notes. Your ticks stay in this browser.

Study strategy

Split your time between the Catalog Explorer interface and SQL practice, roughly one third to two thirds.

In Catalog Explorer, search for tables by a column name and a comment phrase to see how metadata improves discovery. Open a table, add a comment, then add a tag. If your workspace has governed tags configured, notice how the value is restricted to a list. Run a CREATE TABLE AS SELECT from that table and a second query that reads the new table, then open the Lineage tab to see the graph appear and expand it to column level. Read the documentation page on external lineage so you know how outside sources and BI reports are registered.

For SQL, build a ten-row table with deliberate problems: duplicate keys with different timestamps, empty strings, the text 'n/a' in a numeric column, mixed-case country codes and NULLs. Write queries that fix each issue using TRIM, UPPER, NULLIF, TRY_CAST, COALESCE and ROW_NUMBER. Predict each result before running it. Finally, test COUNT(*) against COUNT(column) on a column with NULLs and note the difference, because the exam likes this detail.

FAQ

What share of the Databricks Data Analyst exam is Managing Data?

Managing Data accounts for 8% of the exam, roughly four of the 45 scored questions, according to the official certification page.

What are governed tags?

Governed tags are tags defined at account level with a policy that sets allowed values and who may assign them. They keep classification consistent and can be used by attribute-based access control policies.

What is external lineage?

External lineage records relationships between Unity Catalog assets and systems outside Databricks, such as source applications or BI reports. It is added as external metadata rather than captured automatically.

Which SQL functions matter most for cleaning?

COALESCE, NULLIF, TRY_CAST, TRIM, UPPER and LOWER, plus ROW_NUMBER for deduplication and DELETE or UPDATE for in-place fixes.

Does certifying a table change who can read it?

No. Certification marks a trusted asset. Access is still controlled by Unity Catalog privileges.

Do I need Python for this section?

No. Every objective here can be met with Databricks SQL and the Catalog Explorer interface.

Other Data & Analytics certifications

Where the same subject shows up in other exams. Cards marked "Same subject" open the matching domain.

Test your Domain 2 knowledge

Test whether you can predict cleaning query results, choose governed tags and read lineage correctly on timed questions.

Take the Databricks DA Assoc practice test Check your Databricks DA Assoc readiness

Keep going